Dynamic load balancing method and device for intelligent charging pile management system

Abstract

The invention discloses a dynamic load balancing method and device for an intelligent charging pile management system, and the method comprises the steps: obtaining the operation parameters of a target area transformer corresponding to the intelligent charging pile management system in a current control period, so as to obtain the real-time total load of the target area transformer in the current control period; dynamically calculating the real-time available capacity of the transformer based on the rated power and the real-time total load of the transformer in the target station area; calculating the predicted available capacity of the transformer in the next control period through a load prediction model based on the historical power data and the external environment variables; and according to the real-time available capacity and the predicted available capacity, dynamically adjusting the operation power of the plurality of accessed charging piles, and expanding the number of the accessible charging piles in a load trough period. By fusing the real-time power grid state and the multivariable predicted future load demand, the dynamic fine adjustment of the power of the charging pile and the prospective expansion of the off-peak access amount are realized, the safety of the power grid is ensured, and the capacity utilization efficiency of the transformer is remarkably improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of charging pile control, in particular to a dynamic load balancing method and device for an intelligent charging pile management system. BACKGROUND

[0002] With the rapid increase in the popularity of electric vehicles, the distribution network of old communities is facing increasing pressure from the access of charging infrastructure. Due to the limitations of early planning, the transformer capacity of such communities is generally insufficient, making it difficult to support the simultaneous operation of multiple high-power charging piles. In the prior art, a power collection terminal is usually configured for each charging pile to realize monitoring, which has high hardware cost, large system complexity, and is not conducive to large-scale deployment. Another common solution is to use a time segmentation control strategy based on fixed rules to limit or prohibit the use of charging piles during peak periods. However, this strategy lacks real-time awareness of the state of the power grid and cannot dynamically adjust according to actual load, resulting in underutilization of power grid capacity during low load periods and wasting resources.

[0003] In addition, some centralized charging management systems have been applied, but they mostly rely on traditional power line carrier communication or a single communication mode, which has low transmission rate, weak anti-interference ability, and high response delay. These systems usually lack efficient intelligent scheduling algorithms and real-time data fusion analysis capabilities, and cannot achieve accurate load prediction and dynamic power distribution. More importantly, most existing solutions only focus on real-time load conditions and can only respond passively to overload risks, lacking the ability to predict future load demand and unable to achieve forward-looking charging scheduling and capacity expansion. At the same time, in a multi-phase power supply environment, existing load control methods often ignore the three-phase load balancing problem, which can easily lead to overload in one phase and insufficient load in other phases, reducing system efficiency and potentially causing safety hazards.

[0004] In summary, the existing technology mainly has the problems of high hardware cost, low intelligence level, slow response speed, insufficient resource utilization efficiency, lack of predictive scheduling ability, and three-phase imbalance. SUMMARY

[0005] The purpose of the embodiments of the present application is to provide a dynamic load balancing method and device for an intelligent charging pile management system, which realizes dynamic and fine adjustment of charging pile power and forward-looking expansion of access quantity during low valley period by fusing real-time power grid state and multi-variable predicted future load demand, thereby significantly improving the capacity utilization efficiency of the transformer under the premise of ensuring the safety of the power grid.

[0006] To solve the above technical problems, the first aspect of the embodiments of the present application provides a dynamic load balancing method for an intelligent charging pile management system, comprising the following steps:

[0007] obtaining an operation parameter of a target transformer of a smart charging pile management system in a current control period to obtain a real-time total load of the target transformer in the current control period;

[0008] based on the rated power of the target transformer and the real-time total load, dynamically calculating a real-time available capacity of the transformer;

[0009] based on historical power data and external environmental variables, calculating a predicted available capacity of the transformer in the next control period through a load prediction model;

[0010] based on the real-time available capacity and the predicted available capacity, dynamically adjusting the operation power of a plurality of connected charging piles, and expanding the number of connectable charging piles during the load valley period.

[0011] Further, the calculation of the predicted available capacity of the transformer in the future set period based on the historical power data and the external environmental variables through the load prediction model comprises:

[0012] obtaining a multi-dimensional load prediction feature set, the multi-dimensional load prediction feature set comprising: historical load sequence data, external variable sequence data, and reservation load sequence data;

[0013] through the attention mechanism integrated in the load prediction model, dynamically calculating the weight of each sequence data in the multi-dimensional load prediction feature set and weighted fusion, obtaining a future load sequence of the total load prediction value of the target transformer in the next control period;

[0014] based on the rated power of the transformer and the load prediction value in the future load sequence, the predicted available capacity corresponding to the next control period is calculated.

[0015] Further, the calculation of the predicted available capacity of the transformer in the future set period based on the historical power data and the external environmental variables through the load prediction model comprises:

[0016] normalizing the historical load sequence data, the external variable sequence data and the reservation load sequence data in the multi-dimensional load prediction feature set to obtain standardized multi-source time series data;

[0017] input the standardized multi-source time series data into the attention calculation layer of the load prediction model, and obtain the attention weight distribution of each sequence data at different time steps through the similarity calculation of the query vector and the key vector, wherein the query vector is generated from the hidden state of the current time step, and the key vector is generated from the time step features of each sequence data.

[0018] perform weighted sum fusion processing on the value vectors based on the attention weight distribution to generate a comprehensive feature representation, the value vectors being generated from the feature representations of the sequence data;

[0019] inputting the comprehensive feature representation into a time series prediction layer of the load prediction model for processing to obtain a future load sequence of the predicted value of the total load of the transformer in the target substation in the next control period.

[0020] Further, the historical load sequence data includes: a sequence of total active power values, a sequence of total reactive power values, and a sequence of three-phase current unbalance degrees of the transformer in the target substation within a historical time window;

[0021] The external variable sequence data includes: a sequence of predicted ambient temperature values, a sequence of predicted solar radiation intensity values, and a sequence of date type identifiers, wherein the date type identifiers are used to distinguish weekdays, weekends, and statutory holidays;

[0022] The reservation load sequence data includes: a sequence of expected charging power values of confirmed charging reservations, a sequence of reservation start times, and a sequence of expected charging durations.

[0023] Further, before the real-time available capacity of the transformer is dynamically calculated based on the rated power of the transformer in the target substation and the real-time total load, the method further includes:

[0024] obtaining a multi-dimensional operation evaluation parameter set for dynamically evaluating power grid stability and load fluctuation;

[0025] calculating a dynamic safety power margin adjustment value of the transformer based on the multi-dimensional operation evaluation parameter set;

[0026] calculating the real-time available capacity based on the rated power and the real-time total load in combination with the dynamic safety power margin.

[0027] Further, the multi-dimensional operation evaluation parameter set includes: a power grid frequency deviation value, a three-phase voltage unbalance degree, a total harmonic distortion rate, a load standard deviation, and a maximum mutation rate;

[0028] The calculating of the dynamic safety power margin adjustment value of the transformer based on the multi-dimensional operation evaluation parameter set includes:

[0029] normalizing the power grid frequency deviation value, the three-phase voltage unbalance degree, the total harmonic distortion rate, the load standard deviation, and the maximum mutation rate respectively to obtain corresponding normalized values;

[0030] The normalized value is multiplied by the corresponding preset weight coefficient and then summed by weight to obtain a comprehensive evaluation score that represents the overall operating status of the power grid.

[0031] The comprehensive evaluation score is linearly mapped to a preset percentage range, and the mapping result is used as the dynamic safety power margin adjustment value.

[0032] Wherein, the lower limit of the percentage range is a first preset percentage value of the rated power of the transformer, the upper limit is a second preset percentage value of the rated power of the transformer, and the comprehensive evaluation score is positively correlated with the dynamic safety power margin adjustment value.

[0033] Furthermore, the expansion of the number of accessible charging stations during off-peak periods includes:

[0034] When the predicted available capacity is continuously greater than the real-time available capacity and exceeds a preset threshold, it is determined that the load is in a low period.

[0035] Calculate the maximum number of charging piles that can be expanded to be connected based on the ratio of the current real-time available capacity to the rated power of a single charging pile.

[0036] Send a charging pile availability notification to the user terminal. The availability notification includes the number of available charging piles and the recommended charging time period.

[0037] The system receives a charging access request submitted by the user terminal, allocates the charging pile according to the charging access request, and sends the allocation result to the corresponding user terminal.

[0038] Furthermore, the dynamic adjustment of the operating power of the connected charging piles includes:

[0039] The total power adjustment is determined based on the difference between the real-time available capacity and the predicted available capacity.

[0040] Based on the preset charging pile priority rules, the power adjustment sequence and adjustment amount for each charging pile are determined;

[0041] A control command corresponding to the adjustment amount is generated using pulse width modulation technology;

[0042] The control commands are transmitted to the target charging pile via high-speed power line carrier.

[0043] Accordingly, a second aspect of the present invention provides a dynamic load balancing device for an intelligent charging pile management system, which performs dynamic load balancing processing on intelligent charging piles based on the above-described dynamic load balancing method for an intelligent charging pile management system, including:

[0044] The data acquisition module is used to acquire the operating parameters of the target transformer in the current control cycle of the intelligent charging pile management system, so as to obtain the real-time total load of the target transformer in the current control cycle.

[0045] The data calculation module is used to dynamically calculate the real-time available capacity of the transformer based on the rated power of the transformer in the target area and the real-time total load.

[0046] The prediction calculation module is used to calculate the predicted available capacity of the transformer in the next control cycle based on historical power data and external environmental variables through a load prediction model.

[0047] The device balancing module is used to dynamically adjust the operating power of several connected charging piles based on the real-time available capacity and the predicted available capacity, and to expand the number of connected charging piles during off-peak periods.

[0048] Accordingly, a third aspect of the present invention provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the above-described dynamic load balancing method for a smart charging pile management system.

[0049] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-described dynamic load balancing method for the intelligent charging pile management system.

[0050] The above-described technical solutions of the embodiments of the present invention have the following beneficial technical effects:

[0051] 1. By introducing a forward-looking capacity calculation and dynamic scheduling mechanism based on a multivariate prediction model, the utilization efficiency of power grid resources has been significantly improved. It not only calculates the current available capacity based on real-time transformer load, but more importantly, it integrates historical power data, environmental variables, and user reservation information to accurately predict the load demand and predicted available capacity for future periods through a load prediction model. This enables the system to proactively identify off-peak periods and intelligently expand the number of accessible charging piles while ensuring continuous safety. This solves the problem of resource idleness caused by the lack of prediction capabilities in traditional systems and maximizes the potential of existing power grid infrastructure.

[0052] 2. By monitoring multi-dimensional operating indicators such as grid frequency deviation, voltage imbalance, harmonic distortion rate, and load fluctuation in real time, the system dynamically calculates and adaptively adjusts the safety margin, replacing the traditional conservative strategy of fixed margin. Combined with pulse width modulation technology, the system performs orderly and precise power adjustment on the connected charging piles, forming a closed-loop control system that can sensitively respond to the real-time health status of the grid. This fundamentally enhances the robustness of the system, effectively prevents overload risks, and reduces unnecessary capacity waste.

[0053] 3. By integrating three-phase load balancing optimization functions, the overall stability of the system and the lifespan of equipment are improved; the current and voltage parameters of the three-phase output terminals of the transformer in the distribution area are monitored in real time. Once the imbalance exceeds the threshold, the optimization program is automatically started. The optimal power distribution scheme of each phase is solved by establishing a linear programming model, and precise power redistribution operation is performed to ensure that the three-phase load tends to be balanced. This effectively avoids safety hazards such as low efficiency and equipment overheating caused by single-phase overload, and ensures the stable and efficient operation of the power system. Attached Figure Description

[0054] Figure 1 This is a flowchart of the dynamic load balancing method for the intelligent charging pile management system provided in this embodiment of the invention;

[0055] Figure 2 This is a block diagram of the dynamic load balancing device module of the intelligent charging pile management system provided in an embodiment of the present invention.

[0056] Figure label:

[0057] 1. Data acquisition module; 2. Data calculation module; 3. Predictive calculation module; 4. Equipment balancing module. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments and the accompanying drawings. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0059] Please refer to Figure 1 The first aspect of this invention provides a dynamic load balancing method for an intelligent charging pile management system, comprising the following steps:

[0060] Step S100: Obtain the operating parameters of the target transformer in the current control cycle from the intelligent charging pile management system, so as to obtain the real-time total load of the target transformer in the current control cycle.

[0061] A high-precision smart meter and multi-modal sensor network deployed at the output side of the target transformer and the user's incoming line monitors and acquires key operating parameters such as three-phase voltage, current, power factor, frequency, and harmonic distortion rate in real time at a sampling frequency of no less than once per minute. The collected raw data undergoes preprocessing, including noise suppression based on Kalman filtering and outlier detection and removal (using Z-score or IQR methods). Subsequently, the total active power carried by the transformer in the current control cycle is calculated using an active power integration algorithm. This value serves as the real-time total load characterizing the transformer's real-time load status. Simultaneously, the current data for each phase is recorded for subsequent three-phase unbalance analysis.

[0062] Step S200: Based on the rated power of the target transformer area and the real-time total load, dynamically calculate the real-time available capacity of the transformer.

[0063] First, the pre-stored transformer rated power parameters are retrieved from the local or cloud database, and the real-time total load calculated in step S100 is read. Then, the built-in dynamic safety margin calculation module is accessed. This module performs a multi-dimensional evaluation based on real-time power grid quality indicators (such as frequency deviation, three-phase voltage imbalance, and total harmonic distortion rate) and historical load volatility indicators (such as load standard deviation and maximum mutation rate). A safety margin value between 5% and 10% of the rated power is dynamically generated through a weighted fusion algorithm (such as linear weighting or fuzzy logic). Finally, the power capacity that can be safely used to power charging piles at the current moment is calculated according to the formula "Real-time available capacity = Rated power - Real-time total load - Dynamic safety margin", providing a basis for subsequent power scheduling.

[0064] Step S300: Based on historical power data and external environmental variables, calculate the predicted available capacity of the transformer in the next control cycle using a load prediction model.

[0065] Historical power consumption data (including active power, reactive power, current imbalance, etc.), future weather forecast data (such as temperature, solar radiation intensity), and date type information (weekdays, weekends, holidays) are retrieved from the database. This data is then aggregated with charging reservation requests submitted by users through a mobile application to generate a reserved load sequence containing expected power, start time, and estimated duration. This feature set, composed of multi-source heterogeneous data, is standardized and time-series aligned before being input into a pre-trained multivariate time series prediction model (such as Attention-LSTM or Transformer-based models) with integrated attention mechanisms for inference. The model dynamically fuses the importance of different features through attention weights, outputting a predicted value of the total transformer load for the next control cycle. This prediction is then combined with rated power and dynamic safety margin to calculate the predicted available capacity, providing data support for forward-looking scheduling.

[0066] Step S400: Based on the real-time available capacity and the predicted available capacity, dynamically adjust the operating power of the connected charging piles, and expand the number of connected charging piles during off-peak periods.

[0067] First, the real-time available capacity is compared with the predicted available capacity. If the predicted capacity consistently exceeds the real-time capacity and the difference exceeds a preset threshold (e.g., 10%), it is determined that a period of low load is about to begin. For connected charging piles, duty cycle adjustment commands are generated based on their priority (e.g., priority for public service vehicles and emergency vehicles) using pulse width modulation (PWM) technology. These commands are then transmitted to the target charging pile via high-speed power line carrier (HPLC) or 5G communication to smoothly adjust its output power. Simultaneously, based on the ratio of the current real-time available capacity to the rated power of a single charging pile, the maximum number of charging piles that can be expanded to be connected is calculated. The available number and recommended time periods are then notified to user terminals via mobile applications or SMS interfaces. After receiving charging access requests submitted by users, resource allocation and conflict detection are performed. Finally, all parameters are updated according to the new load status, and a closed-loop control mechanism is activated to ensure stable operation.

[0068] Specifically, step S300, which calculates the predicted available capacity of the transformer within a future set time period based on historical power data and external environmental variables using a load forecasting model, includes:

[0069] Step S310: Obtain a multi-dimensional load prediction feature set, which includes: historical load sequence data, external variable sequence data, and scheduled load sequence data.

[0070] A multi-dimensional load prediction feature set is obtained from the system database or external data interface. This feature set consists of historical load sequence data, external variable sequence data, and scheduled load sequence data. Historical load sequence data covers time-series data such as total active power, total reactive power, phase current, and power factor of the target transformer area within a set time window, typically at the minute or hourly granularity. External variable sequence data includes weather forecasts for the next few hours or days (such as ambient temperature, humidity, and solar radiation intensity), date types (such as weekdays, weekends, and public holidays), and special event markers (such as large-scale events and extreme weather warnings). Scheduled load sequence data originates from charging requests submitted by users through mobile applications or the web, including desired charging power, start time, duration, and charging station location. Before being input into the model, this multi-source data undergoes rigorous cleaning, alignment, and normalization. For example, the Z-score method is used to standardize numerical features, and one-hot encoding is used for categorical features to eliminate dimensional differences and improve model training stability.

[0071] Step S320: Through the attention mechanism integrated in the load prediction model, the weights of each sequence data in the multi-dimensional load prediction feature set are dynamically calculated and weighted and fused to obtain the future load sequence of the total load prediction value of the target transformer area in the next control cycle.

[0072] By integrating an attention mechanism into the load forecasting model, the weights of each sequence data in the multi-dimensional load forecasting feature set at different time steps are dynamically calculated and weighted fused to generate the future load sequence of the total load forecast value of the target transformer area in the next control cycle. Specifically, the model adopts an encoder-decoder architecture, where the encoder maps the input multi-source time series into hidden representations, and the decoder dynamically selects the historical information most relevant to the current forecast based on the attention mechanism. The calculation of attention weights depends on the similarity between the query (usually the decoder's hidden state), key, and value (both are hidden representations output by the encoder), and is commonly implemented using scaled dot product attention. The weighted fused context vector and the decoder state are input together to a fully connected layer or a temporal convolutional layer, ultimately outputting the load forecast values ​​for multiple future time points. This mechanism enables the model to adaptively focus on the importance of different features and historical periods, such as increasing the weight of temperature features during hot weather or strengthening the influence of date types during holidays, thereby significantly improving prediction accuracy and robustness.

[0073] Step S330: Based on the transformer's rated power and the predicted load values ​​in the future load sequence, calculate the predicted available capacity corresponding to the next control cycle.

[0074] First, the rated power parameters of the transformer are read from the configuration file, and combined with the future load sequence output in step S320, the predicted load value is calculated point by point in time. Then, the dynamic safety margin calculation module is invoked. This module comprehensively considers real-time grid quality indicators (such as frequency deviation, voltage fluctuation, and harmonic distortion rate) and load change trends to generate a safety margin value that adjusts over time, usually expressed as a percentage of rated power. Finally, the estimated available capacity at each future time point is obtained using the formula "Predicted Available Capacity = Rated Power - Predicted Load - Dynamic Safety Margin". This capacity value not only reflects the future power margin of the transformer but also provides a crucial basis for advance decisions on charging scheduling, power allocation, and capacity expansion. For example, it allows for advance notification of available charging resources to users when a low load period is predicted, or proactive reduction of operating power before peak load periods to avoid overload.

[0075] Further, in step S320, the weights of each sequence data in the multi-dimensional load prediction feature set are dynamically calculated and weighted fused using the attention mechanism integrated in the load prediction model to obtain the future load sequence of the total load prediction value of the target transformer area in the next control cycle, including:

[0076] Step S321: Normalize the historical load sequence data, external variable sequence data, and scheduled load sequence data in the multi-dimensional load prediction feature set to obtain standardized multi-source time series data.

[0077] This step is the core of the data preparation phase, directly impacting the subsequent model training performance and prediction accuracy. The preprocessing process begins with historical load series data, which typically includes time-series indicators such as total active power, total reactive power, phase current, and power factor. Because these indicators have different dimensions and may exhibit abnormal fluctuations, Z-score normalization is used. This involves subtracting the mean of the series from each data point and dividing by the standard deviation, resulting in standardized data with a mean of 0 and a standard deviation of 1. This processing not only eliminates dimensional differences but also effectively suppresses the influence of outliers caused by measurement errors or transient interference through statistical methods, improving data quality and consistency.

[0078] Preprocessing of external variable sequence data requires distinguishing between continuous and categorical features. Continuous features, such as ambient temperature, humidity, and solar radiation intensity, are linearly transformed to the [0,1] interval using the Min-Max scaling method to preserve the original distribution while adapting to the numerical range requirements of the model input. Categorical features, such as date type (weekday, weekend, holiday) and weather conditions (sunny, rainy, snowy, etc.), are converted into numerical representations through one-hot encoding, with each category mapped to a binary vector for easy model recognition and processing. Reservation load sequence data originates from user-submitted charging requests and includes information such as expected power, start time, and duration. This type of data requires time-series alignment to ensure consistency with the timestamps of other data sequences, and any missing values ​​are filled using forward imputation or linear interpolation to guarantee data continuity and integrity. All preprocessed multi-source time series data are ultimately integrated into a unified-dimensional tensor structure. This structured representation facilitates parallel processing and feature extraction in subsequent models while effectively preserving the temporal correlation and statistical characteristics of the original sequences.

[0079] Step S322: Input the standardized multi-source time series data into the attention calculation layer of the load prediction model. By calculating the similarity between the query vector and the key vector, obtain the attention weight distribution of each sequence data at different time steps. The query vector is generated from the hidden state of the current time step, and the key vector is generated from the time step features of each sequence data.

[0080] The attention layer of a load prediction model typically employs a multi-head self-attention mechanism or a cross-attention mechanism to dynamically allocate weights, serving as a crucial module for feature selection and fusion. Technically, the query vector is generated from the hidden state of the decoder at the current time step through a linear transformation, while the key vector is obtained by mapping the time-step features of each sequence data output by the encoder through the same linear layer. This shared parameter design helps maintain the consistency of the feature space. Similarity calculation uses the classic scaled dot product attention algorithm, which calculates the dot product of the query vector and the key vector and divides it by the square root of the vector dimension. This scaling operation effectively avoids the gradient vanishing problem caused by excessively large dot product values, enhancing training stability. The calculation results are then normalized using the Softmax function to obtain the attention weight distribution of each sequence data at different time steps.

[0081] This computational process enables the model to dynamically capture the correlations of different feature sequences at different time steps, demonstrating the core advantage of the attention mechanism. For example, when predicting load during the evening peak electricity consumption period, the model automatically assigns higher attention weights to historical load data from the same period, thereby better capturing the daily cyclical variation patterns. When meteorological data shows an impending cold wave, the influence of temperature sequences is amplified, making the prediction results more reflective of the specific impact of extreme weather on electricity load. This adaptive weight allocation mechanism greatly enhances the model's ability to understand complex multi-source data, laying a solid foundation for subsequent feature fusion.

[0082] Step S323: Based on the attention weight distribution, the value vector is weighted and fused to generate a comprehensive feature representation. The value vector is generated from the feature representations of each sequence data.

[0083] Based on the attention weight distribution calculated in step S322, the value vectors are weighted and summed to generate a comprehensive feature representation. The value vectors are generated from the feature representations of each sequence data through independent linear transformations, maintaining the same dimension as the key vectors. This design ensures the matching of the feature space. During the weighted summation process, the value vector at each time step is multiplied by its corresponding attention weight and then summed to form a context-aware feature vector. This operation is essentially a recalibration and fusion of multi-source feature information, where the attention weights act as a "feature selector," highlighting the information components most relevant to the current prediction task.

[0084] The generated comprehensive feature representation not only aggregates global information from multi-source sequences but also highlights key feature components through attention weights. For example, in summer forecasting, this representation may integrate the collaborative representation of multiple factors such as historical load trends, temperature effects related to air conditioning use, and load increases due to evening charging reservations; while in holiday forecasting, it strengthens the influence of date type features and reduces the weight of weekday load patterns. This comprehensive feature representation, rich in semantic and relational information, provides high-quality input to the time series prediction layer, greatly improving the model's feature representation capability and prediction accuracy.

[0085] Step S324: Input the comprehensive feature representation into the time-series prediction layer of the load prediction model for processing to obtain the future load sequence of the total load prediction value of the target transformer area in the next control cycle.

[0086] The comprehensive feature representation is input into the temporal prediction layer of the load prediction model for processing. This layer typically consists of stacked Long Short-Term Memory (LSTM) units, Gated Recurrent Units (GRUs), or Temporal Convolutional Networks (TCNs), responsible for capturing long-term dependencies in the sequence and generating future load sequences. These network structures each have their advantages: LSTMs effectively address the long-term dependency problem through a sophisticated gating mechanism; GRUs simplify the structure and improve computational efficiency while maintaining similar performance; and TCNs utilize dilated convolutions to expand the receptive field, resulting in higher parallelism. In practical applications, appropriate network structures can be selected or combined according to specific needs.

[0087] The comprehensive feature representation is first dimensionally adjusted through a fully connected layer to match the input requirements of the prediction layer, and then fed into a recurrent neural network for multi-step forward prediction. The decoder employs an autoregressive structure, using the prediction output of the current time step as the input for the next time step, progressively generating the entire prediction sequence. Although this recursive prediction method is computationally intensive, it maintains good temporal consistency of the sequence. The final output is a future load sequence of the predicted total load of the target transformer area within the next control cycle. Its length and granularity are determined by the control cycle setting, such as 96 predicted values ​​at 15-minute intervals for the next 24 hours, or 168 predicted values ​​at hourly intervals for the next week. This sequence accurately reflects the load change trend of the transformer in the future, including daily cycle patterns, weekly cycle patterns, and the impact of special events, providing core data support for system capacity assessment and scheduling decisions.

[0088] Through the synergistic effect of the aforementioned multi-layer attention mechanism and time-series prediction model, key information from multi-source heterogeneous data can be adaptively fused, significantly improving the accuracy and robustness of load prediction. The dynamic weight allocation mechanism effectively solves the problem that traditional models are susceptible to noise interference and feature redundancy in long-sequence prediction, enhancing the model's adaptability to the complex operating environment of the power grid. The resulting high-precision future load sequence provides a reliable input for calculating the predicted available capacity of transformers, thereby enabling forward-looking charging scheduling and resource optimization, maximizing asset utilization efficiency and user satisfaction while ensuring the safe and stable operation of the power grid.

[0089] Furthermore, in a specific embodiment of the present invention, the historical load sequence data includes: the total active power value sequence, the total reactive power value sequence, and the three-phase current imbalance sequence of the target transformer within the historical time window.

[0090] Historical load series data serves as the fundamental input for load forecasting models, capturing the historical operating patterns and load variation trends of transformers in the target distribution area. This dataset specifically includes: a total active power value series, which continuously records the active power output of transformers within a specific historical time window (e.g., the past 7 days) at fixed sampling intervals (e.g., 15 minutes), directly reflecting the actual energy consumption load of the transformers; a total reactive power value series, which records reactive power changes within the same time window, crucial for assessing the system's power factor, reactive power compensation requirements, and the overall operating efficiency of the power grid; and a three-phase current imbalance series, which quantifies the balance of three-phase load by calculating the degree to which the maximum current deviates from the average current (usually expressed as a percentage), a key indicator for identifying phase-to-phase load differences, preventing single-phase overloads, and assessing potential thermal stress risks in equipment. This data is typically acquired at high frequencies using advanced metering systems (AMI) smart meters, power quality analyzers, and other equipment installed on the transformer side. After preprocessing such as data cleaning, outlier removal, and time alignment, it is stored in a time-series database, providing high-quality training and inference data for the model.

[0091] The external variable sequence data includes: the predicted environmental temperature sequence for the next control period, the predicted solar radiation intensity sequence, and the date type identifier sequence, where the date type identifier is used to distinguish between weekdays, weekends, and statutory holidays.

[0092] The external variable sequence data incorporates external environmental and human factors affecting electricity load, aiming to improve the accuracy and generalization ability of the load forecasting model. This data includes: an ambient temperature forecast sequence, sourced from the forecast service interface of authoritative meteorological departments, providing accurate temperature forecasts at hourly intervals for the next control period (e.g., the next 24 hours), as ambient temperature is strongly correlated with the electricity consumption of temperature-sensitive loads such as air conditioning and heating; a solar radiation intensity forecast sequence, also derived from meteorological forecasts, whose data (usually in watts per square meter) not only affects the cooling load of buildings (through solar radiation heat gain) but also directly determines the potential output power of photovoltaic power generation systems, thus indirectly affecting the net load; and a date type identifier sequence, a categorical variable sequence used to label the type of each day in the next control period (e.g., ordinary workday, weekend, national statutory holiday). This identifier is derived from publicly available calendar services and can effectively capture the significant differences in human social activity patterns and industrial and commercial operation rhythms under different date types and their systematic impact on electricity consumption habits.

[0093] The scheduled load sequence data includes: the sequence of expected charging power values ​​for confirmed charging reservations, the sequence of reservation start times, and the sequence of estimated charging durations.

[0094] The scheduled load sequence data is derived directly from users' charging reservation behavior, enabling the quantification and integration of future deterministic load demands. This dataset consists of the following components: a sequence of expected charging power values ​​for confirmed charging reservations, summarizing user-specified or system-recommended charging power values ​​from all confirmed charging requests through the platform; a reservation start time sequence, precisely recording the planned charging start time for each reservation request (typically accurate to the minute); and an estimated charging duration sequence, calculated based on the user-inputted expected charging amount or state of charge (SOC) target value, vehicle battery parameters, and charging power. By fusing these three types of sequences, the load curve profile generated by electric vehicle charging activities in the next control cycle can be proactively constructed, transforming the originally uncertain random charging demand into a relatively deterministic future load increment that can be incorporated into the predictive model. This is a crucial prerequisite for achieving accurate prediction and proactive scheduling.

[0095] Furthermore, before dynamically calculating the real-time available capacity of the transformer based on the rated power and real-time total load of the target distribution area transformer in step S400, the following steps are also included:

[0096] Step S401: Obtain a multi-dimensional operational evaluation parameter set for dynamically assessing grid stability and load volatility.

[0097] To move beyond simple static power calculations and achieve truly safe and reliable dynamic load balancing, it is first necessary to obtain a multi-dimensional set of operational evaluation parameters that comprehensively reflects the real-time health of the power grid. This parameter set is the fundamental data source for advanced decision-making. This set includes not only traditional power parameters but also indicators such as power quality and load fluctuation characteristics. Specifically, these parameters include: grid frequency deviation, which reflects the instantaneous balance between power generation and consumption; a frequency deviation from the rated value (e.g., 50Hz) is an important early indicator of grid stress; three-phase voltage imbalance, which assesses the phase-to-phase balance of the distribution system by measuring the difference in the amplitude of the three-phase voltages; severe imbalance can lead to decreased transformer efficiency and overheating; total harmonic distortion (THD), which quantifies the degree to which the current or voltage waveform in the grid deviates from the standard sine wave; high harmonic content can exacerbate equipment losses and affect the normal operation of sensitive equipment; load standard deviation, a statistic calculated based on recent load data (e.g., the past 15 minutes), used to quantify the severity of load fluctuations around the mean, characterizing load stability; and maximum abrupt change rate, which captures the maximum instantaneous rate of change of the load within a specific time window, used to identify sharp increases or decreases in load, which places demands on the response speed of the control system. These parameters are sampled at high speed and calculated in real time by intelligent sensors and power quality monitoring terminals deployed at the transformer outlets of the distribution area, and then aggregated to the central management system.

[0098] Step S402: Calculate the dynamic safety power margin adjustment value of the transformer based on the multi-dimensional operation evaluation parameter set.

[0099] After obtaining multi-dimensional operational evaluation parameters, these heterogeneous parameters with different dimensions need to be integrated into a unified and operable decision indicator, namely the dynamic safety power margin adjustment value. This process is a multi-factor comprehensive evaluation process. First, each obtained raw parameter value is normalized, mapping it linearly or non-linearly to the interval [0,1]. The design of the normalization function considers the safe operating threshold of each parameter. For example, when the frequency deviation is within the allowable range of the national standard, its normalized value may be close to 0, indicating a good condition; while when it is close to or exceeds the limit, the normalized value will approach 1, indicating a poor condition.

[0100] Subsequently, each normalized parameter value is multiplied by a pre-defined weighting coefficient reflecting its importance. These weighting coefficients are typically determined by domain expert knowledge or machine learning algorithms analyzing historical fault data; for example, in areas with poor voltage stability, the weight of three-phase voltage imbalance may be set higher. All weighted values ​​are summed to produce a comprehensive evaluation score between 0 and 1. This score is a dimensionless value; the higher the value, the more strained and risky the current overall operation of the power grid.

[0101] Finally, this comprehensive evaluation score is transformed into a preset percentage range using a linear mapping function. The lower limit of this range is typically a small percentage (e.g., 5%) of the transformer's rated power, while the upper limit is a larger percentage set based on conservative principles (e.g., 20%). The mapping relationship is designed to be positively correlated, meaning that the higher the comprehensive evaluation score (the worse the grid condition), the larger the calculated dynamic safety power margin adjustment value. This implies that when the grid condition is assessed as poor, a larger safety buffer will be proactively reserved when calculating available capacity, sacrificing some utilization for higher operational safety and reliability.

[0102] Step S403: Calculate the real-time available capacity based on the rated power and real-time total load, combined with the dynamic safety power margin.

[0103] After calculating the dynamic safety margin adjustment value, the final calculation of real-time available capacity is performed. This process combines the transformer's physical limits, the current actual load, and the aforementioned calculated dynamic safety requirements. The transformer's rated power (S) r The apparent power capacity (L) is a fixed physical upper limit value, representing the maximum apparent power capacity allowed for long-term operation. Real-time total load (L) now The current total apparent power is obtained in real time through the measurement system.

[0104] Traditional static calculation methods may simply use (S r -L now The actual available capacity is calculated as the difference between the calculated and actual power margins. However, this invention employs a more advanced and secure dynamic calculation method. The core idea is that the true "available" capacity is not a simple arithmetic difference, but rather a dynamic safety buffer space reserved to address grid uncertainties, fluctuations, and potential risks. This safety buffer space is the dynamic safety power margin adjustment value (ΔS) calculated in the previous step. safe ).

[0105] Therefore, real-time available capacity (C available The calculation logic for C can be expressed as follows: subtract the current real-time total load from the transformer's rated power, and then further subtract the dynamic safety power margin calculated based on the real-time operational health of the power grid. That is: C available =S r -L now -ΔS safe For example, a transformer with a rated power of 800kVA has a current real-time load of 500kVA. If the current power grid condition is assessed as excellent and very stable, then ΔS safeIt might only be 5% of the rated power (40kVA), then the real-time available capacity is 260kVA; if large fluctuations and harmonic pollution are detected in the power grid, ΔS is calculated after comprehensive evaluation. safe If the rated power is 15% (120kVA), then even if the real-time load is still 500kVA, the real-time available capacity will be dynamically adjusted to 180kVA. This method ensures that the necessary safety margin is reserved for control decisions under any operating condition, greatly enhancing robustness and preventative control capabilities.

[0106] In one specific embodiment of the present invention, the multi-dimensional operation evaluation parameter set includes: grid frequency deviation, three-phase voltage imbalance, total harmonic distortion rate, load standard deviation, and maximum mutation rate.

[0107] Specifically, the grid frequency deviation refers to the instantaneous difference between the actual operating frequency of the power grid and the rated frequency (such as 50Hz or 60Hz). This parameter is a core indicator reflecting the real-time balance between power generation and consumption. Frequency deviation is obtained by high-frequency sampling (e.g., 100 times per second) of the voltage cycle at the transformer's grid connection point and calculating its accuracy, typically reaching the millihertz level. When the load suddenly increases and the power generation fails to keep up, the frequency will decrease; conversely, a sudden drop in load will cause the frequency to increase. Therefore, the frequency deviation can sensitively indicate the stress level and stability of the regional power grid. A small and stable frequency deviation indicates that the grid is in a good balance; while a sustained or large frequency deviation indicates a risk of power shortage or surplus, which may trigger protective actions such as low-frequency load shedding or high-frequency generator tripping. Frequency deviation is considered an important dynamic assessment parameter; the larger the value (regardless of whether positive or negative), the higher the perceived risk to grid stability, thus requiring consideration when calculating safety margins.

[0108] Specifically, three-phase voltage imbalance is an indicator that measures the degree of symmetry of the three-phase voltage amplitude in a power distribution system, usually expressed as the percentage of the negative-sequence voltage component to the positive-sequence voltage component. This parameter is measured by a power quality monitoring device installed at the low-voltage side outlet of the transformer. Imbalance mainly stems from the uneven distribution of single-phase loads (such as residential electricity consumption and some charging piles) among the three phases. Severe imbalance can cause multiple hazards to the power grid and equipment: First, it leads to additional heating in the transformer windings, reducing its actual load-carrying capacity to below the rated capacity, the so-called "derating operation"; second, it causes a decrease in motor efficiency and additional vibration, and may cause non-characteristic subharmonics in power electronic equipment. Therefore, three-phase voltage imbalance is directly related to the thermal safety of the transformer itself and the power supply quality of the entire power distribution system. This system incorporates it into its evaluation framework. Once the imbalance exceeds a preset threshold, it automatically increases the safety margin adjustment value, thereby adopting a more conservative strategy when allocating charging power to avoid exacerbating the imbalance or triggering protection devices.

[0109] Specifically, Total Harmonic Distortion (THD) is a comprehensive indicator that quantifies the degree to which the voltage or current waveform of a power grid deviates from an ideal sine wave. It is defined as the percentage of the effective value of all harmonic components to the effective value of the fundamental component. Modern charging piles, as typical nonlinear power electronic loads, are a significant source of power grid harmonics. High-order harmonic currents generate harmonic voltage drops across the power grid impedance, causing voltage waveform distortion. The hazards of harmonics are widespread, including increasing skin effect losses and core losses in transformers and lines, leading to equipment overheating and reduced efficiency; potentially causing resonant overloads in power capacitors; and interfering with the normal operation of precision measurement and control equipment. THD data is obtained by power quality analyzers through Fast Fourier Transform analysis of current and voltage waveforms. A high THD indicates severe power grid "pollution" and a deteriorated electrical environment. By monitoring this parameter, when harmonic content is high, it indicates increased electrical stress, thus allowing for a greater safety margin to address the additional thermal effects and potential equipment failure risks caused by harmonics.

[0110] Specifically, the load standard deviation is a statistical indicator used to quantify the magnitude of fluctuations in the total transformer load around its average value within a specific time window (e.g., the past 15 minutes); it calculates the dispersion of the load value at each sampling point from the average load. A smaller load standard deviation indicates a smooth, slow, and predictable load curve; while a larger load standard deviation means the load is experiencing sharp and frequent fluctuations. This volatility, especially random fluctuations that cannot be predicted by historical patterns or pre-set information, poses a challenge to the response speed and stability of the control system. Frequent power surges can quickly invalidate available capacity calculations based on earlier data, increasing the risk of overload. Therefore, the load standard deviation parameter provides a quantitative view of load stability. When a significant increase in current load volatility is detected, even if the instantaneous absolute load value is not high, a dynamic safety margin will be appropriately increased as a precaution, providing a buffer to cope with possible sudden load increases and enhancing the robustness of the system.

[0111] The maximum change rate parameter captures the absolute value of the maximum rate of change of load power within a set observation time interval (e.g., 1 minute), typically measured in kilowatts per minute (kW / min). It directly reflects the severity and abruptness of load changes. For example, the simultaneous start-up and shutdown of several high-power charging piles or the switching on and off of large industrial equipment can generate extremely high load change rates. The impact of such abrupt changes on the power grid is far greater than that of slow, equal-volume load changes, requiring transformers, protection devices, and control systems to have extremely fast response capabilities. A high maximum change rate means that the system state may change instantaneously, leaving a very short window for the control system to make decisions and respond. This parameter is introduced to identify such extreme dynamic conditions. Once the load change rate is detected to exceed the normal range, it immediately indicates that a high-risk, high-uncertainty transient process is in progress, thus significantly raising the safety alert level. The dynamic safety margin adjustment value will be increased accordingly to ensure safe and stable operation in the most conservative way during any rapid changes, preventing overload events caused by response delays.

[0112] Accordingly, step S402, based on a multi-dimensional operational evaluation parameter set, calculates the dynamic safety power margin adjustment value of the transformer, including:

[0113] Step S4021: Normalize the grid frequency deviation, three-phase voltage imbalance, total harmonic distortion, load standard deviation and maximum mutation rate to obtain the corresponding normalized values.

[0114] Transforming heterogeneous, multi-dimensional operational evaluation parameters with different dimensions into standardized values ​​that can be comprehensively calculated is the first step in calculating dynamic safety power margin. The purpose of normalization is to eliminate the influence of differences in dimensions and orders of magnitude among the parameters and to map them uniformly to the same numerical range, typically [0,1]. Normalization functions based on national standards, industry specifications, or historical operational data are preset for each type of parameter. For grid frequency deviation, the normalization function is usually designed with the rated frequency (e.g., 50.00Hz) as the center; the larger the absolute value of the frequency deviation, the higher the normalized value, reflecting greater grid balance pressure. When dealing with three-phase voltage imbalance, a piecewise function is used for mapping according to the limits specified in the GB / T15543 national standard for power quality (e.g., 2%): within the allowable range of the national standard, the normalized value is low and increases slowly; once the limit is exceeded, the normalized value will rise sharply to highlight the high risk it brings. The total harmonic distortion rate (THD) is handled similarly, with its normalization function constructed based on the limits for the corresponding voltage level in the GB / T 14549 Power Quality Public Power Grid Harmonic Standard. For the two statistical parameters, load standard deviation and maximum mutation rate, the normalization benchmark is more dynamic, typically set based on the distribution characteristics of the transformer's recent (e.g., the past week) historical data (e.g., maximum value, average value, or specific percentile). Through this carefully designed normalization system, parameters from different physical domains are transformed into comparable, dimensionless scaling values, laying a solid foundation for subsequent comprehensive evaluation.

[0115] Step S4022: Multiply the normalized value by the corresponding preset weight coefficient and then perform a weighted summation to obtain the comprehensive evaluation score that represents the overall operating status of the power grid.

[0116] After obtaining the normalized values ​​of each parameter, the next step is to integrate these scattered indicators into a single score that comprehensively represents the current overall operating status of the power grid. This process employs a weighted summation algorithm. A preset weight coefficient is assigned to each normalized parameter; these weight coefficients are crucial, determining the relative importance of each parameter in the final evaluation. The weight assignments are not arbitrary but deeply integrate professional knowledge of power system operation with historical data analysis results. For example, grid frequency deviation directly relates to the stability and security of the regional power grid, and its weight is usually assigned a high value; three-phase voltage imbalance directly affects the load-carrying capacity and lifespan of transformers, and its weight is also significant; total harmonic distortion reflects power quality, and its weight is secondary; while load standard deviation and maximum mutation rate describe the fluctuation characteristics of the load, their weights are relatively small but indispensable, especially in scenarios with frequent load changes. During calculation, the normalized value of each parameter is multiplied by its corresponding weight coefficient, and all multiplications are summed to obtain a comprehensive evaluation score between 0 and 1. This score is a highly condensed indicator. The closer the score is to 0, the better and more stable the overall operation of the power grid. The closer the score is to 1, the more tense the power grid is, and the more prominent the problems of volatility, imbalance and harmonic pollution are, and the higher the overall operational risk.

[0117] Step S4023: Linearly map the comprehensive evaluation score to a preset percentage range, and use the mapping result as the dynamic safety power margin adjustment value.

[0118] After obtaining the comprehensive evaluation score, it needs to be converted into a control parameter with clear physical meaning and operational guidance value—the dynamic safety power margin adjustment value. This adjustment value represents the additional power buffer space recommended to address the current comprehensive risks of the power grid. This conversion is accomplished through a linear mapping function. A percentage range is preset, with both the lower limit (L) and upper limit (U) being fixed proportions of the transformer's rated power. For example, the lower limit L can be set to 2% of the rated power, representing the minimum safety margin that must be maintained even under excellent grid conditions; the upper limit U can be set to 15% of the rated power, representing the maximum safety margin that needs to be maintained under extremely deteriorating grid conditions. The mapping function is designed to have a positive correlation, meaning that the higher the comprehensive evaluation score (S), the higher the calculated dynamic safety power margin adjustment value (ΔS). safe The larger ΔS is, the greater its calculation logic can be expressed as: ΔS safe = L + S * (UL). This means that when S = 0 (ideal state), ΔS safe Take the minimum value L; when S = 1 (worst-case scenario), ΔS safe Take the maximum value U; when in an intermediate state, use linear interpolation proportionally. Ultimately, ΔS safeAs a power value (unit: kW or kVA) directly used for capacity calculation, it accurately quantifies the abstract and comprehensive risk assessment results into a specific and executable power control command, thereby realizing a closed loop from "state perception" to "decision output".

[0119] The lower limit of the percentage range is the first preset percentage value of the transformer's rated power (which can be 5%), and the upper limit is the second preset percentage value of the transformer's rated power (which can be 10%). The comprehensive evaluation score is positively correlated with the dynamic safety power margin adjustment value.

[0120] Furthermore, the dynamic adjustment of the operating power of the connected charging piles in step S400 includes:

[0121] Step S411: Determine the total power adjustment amount based on the difference between the real-time available capacity and the predicted available capacity.

[0122] By comparing the difference between real-time available capacity and predicted available capacity, the total amount of power adjustment required for the connected charging pile cluster is determined. The core idea of ​​this step is to use the foresight of predictive information to smooth load changes and avoid responding only to potentially drastic power fluctuations caused by real-time load. Real-time available capacity reflects the transformer's actual instantaneous availability, while predicted available capacity is an estimated value of the available capacity for the next control cycle, derived from a load forecasting model.

[0123] When the predicted available capacity consistently exceeds the real-time available capacity and surpasses a certain set threshold, it indicates a transition from peak load to off-peak load, with the transformer having increasing idle capacity. At this point, a positive power regulation can be calculated. This regulation is equal to the difference between the predicted and real-time available capacity, but does not exceed a maximum permissible upward adjustment limit set based on historical data and safety considerations. This positive regulation means the system can issue a command to allow the charging station to appropriately increase its operating power, accelerating the charging process.

[0124] Conversely, when the predicted available capacity is less than the real-time available capacity, it indicates that the load may be about to increase or a sudden load event may occur, and the available transformer capacity will decrease. In this case, a negative power regulation value is calculated, which is equal to this difference. This negative regulation value means that an instruction needs to be issued requiring the entire charging pile cluster to reduce its operating power to reserve the necessary safety capacity for the upcoming load increase and prevent transformer overload.

[0125] Step S412: Based on the preset charging pile priority rules, determine the power adjustment sequence and adjustment amount for each charging pile.

[0126] After determining the total power adjustment amount, this total needs to be reasonably decomposed and allocated to each specific charging station. This is a resource allocation decision-making process that requires comprehensive consideration of multiple factors and is executed according to a set of preset, configurable priority rules. These rules typically form a multi-level decision tree to ensure the fairness and efficiency of the allocation and to meet specific operational objectives.

[0127] The highest priority is usually given to charging stations that are in critical service status, such as those performing emergency charging tasks (e.g., ambulances, fire trucks) or vehicles with extremely low battery levels (e.g., SOC below 10%). The power requirements of these stations are usually prioritized, and their power may not be reduced or reduced only slightly.

[0128] Secondly, the user's charging service contract level will be considered. Users who have signed priority service or high-rate contracts with operators will have higher priority in power allocation for their charging stations. Their charging power will be increased earlier than that of users with ordinary contracts, even if the latter are downgraded later.

[0129] Then, the overall charging efficiency will be optimized. For charging piles in the constant power stage with high charging efficiency, maintaining or increasing their power appropriately helps to complete charging faster and free up capacity; while for piles that have entered the end of trickle charging, reducing their power has little impact on the total charging time, but can effectively release capacity.

[0130] Finally, to ensure fairness, under similar conditions, a polling mechanism can be used to ensure that the power adjustment targets are not always the same batch of charging piles, thus preventing some users from being adversely affected in the long term.

[0131] Step S413 generates control commands corresponding to the adjustment amount through pulse width modulation technology.

[0132] To achieve precise, smooth, and stepless control of the charging pile's output power, pulse width modulation (PWM) technology is used to generate the final control command. PWM technology is an effective method to simulate different average output power by controlling the duty cycle of a fixed-frequency pulse signal (i.e., the ratio of the high-level time within one cycle to the entire cycle).

[0133] The system's main controller calculates the target adjustment power for each charging station and then converts this into the target duty cycle of the required PWM control signal. This conversion process is typically based on the characteristic curves of the power devices inside the charging station or a preset mapping table. For example, a target power of 70% of the rated power might correspond to a PWM wave with a 70% duty cycle.

[0134] The task of generating the PWM wave is accomplished by a dedicated timer / counter hardware module within the main controller. This module can generate highly accurate and stable digital pulse signals, the frequency of which and the duty cycle can be precisely set via software registers. Hardware generation ensures the timing accuracy and reliability of the control signals, avoiding delays and jitter caused by software loops.

[0135] Step S414 sends control commands to the target charging pile via high-speed power line carrier.

[0136] After generating control commands, they need to be reliably and quickly transmitted to the target charging stations scattered in various locations. High-speed power line carrier communication was chosen as the transmission method. This is a data transmission technology that uses existing power lines as the communication medium, eliminating the need for additional communication lines and greatly reducing deployment costs and complexity.

[0137] At the transmitting end, the main station modulation equipment located near the transformer in the distribution area modulates the digital code stream of the PWM control command generated in step S413 onto a high-frequency carrier signal of 2MHz to 30MHz using modern modulation techniques such as orthogonal frequency division multiplexing. This modulated high-frequency signal is then injected into the low-voltage power line through a coupling circuit and transmitted on the power line together with the 50Hz power frequency current.

[0138] HPLC technology features high bandwidth and good real-time performance, supporting millisecond-level command issuance. The data frame structure is carefully designed, including a unique identifier for the target charging station, encrypted control command content, and error check codes, ensuring that commands are accurately delivered to the designated charging station. Furthermore, if an error occurs during transmission, the receiving end can identify and discard the data, requesting a retransmission.

[0139] At the receiving end, each smart charging pile is equipped with an HPLC demodulation module. This module separates and reconstructs the digital control commands from the transmitting end through filtering and demodulation. The charging pile's main control chip parses these commands, obtains the target power or duty cycle information, and then drives its internal power conversion circuits (such as AC / DC rectifiers and DC / DC converters) to adjust the power output to the electric vehicle. Ultimately, it precisely executes the power adjustment commands issued by the system, completing the entire closed-loop control.

[0140] Furthermore, step S400, which involves expanding the number of accessible charging stations during off-peak hours, includes:

[0141] Step S421: When the predicted available capacity is continuously greater than the real-time available capacity and exceeds the preset threshold, it is determined that the load is in a low period.

[0142] The determination of load trough periods is not based on a single instantaneous value comparison, but rather introduces a dual judgment mechanism that includes duration and amplitude thresholds to prevent misjudgments caused by instantaneous load fluctuations. Two key indicators, predicted available capacity and real-time available capacity, are continuously monitored and compared. Predicted available capacity is an estimated value of transformer available capacity over a future period based on a load forecasting model, reflecting a prediction of future power supply and demand; real-time available capacity reflects the actual available capacity of the transformer at the current moment.

[0143] When the predicted available capacity is consistently greater than the real-time available capacity, and the difference exceeds a preset power threshold (e.g., 100kW), a preliminary judgment condition is triggered. However, meeting this condition alone does not immediately confirm a load off-peak period; it is also necessary to confirm that the state has been stably maintained for a preset time window (e.g., 15 minutes). This time window is designed to filter out short-lived, random load fluctuations, ensuring that the identified trend has sufficient stability and reliability. Only when both conditions—"the difference exceeds the threshold" and "the state persists for a period of time"—are met simultaneously will the power grid be officially determined to have entered a load off-peak period. This rigorous judgment logic significantly improves the accuracy of state identification, providing a reliable premise for subsequent capacity expansion decisions.

[0144] Step S422: Calculate the maximum number of charging piles that can be expanded to be connected based on the ratio of the current real-time available capacity to the rated power of a single charging pile.

[0145] Once the off-peak load period is confirmed, it's crucial to accurately calculate how many additional charging stations can be safely connected. The core basis for this calculation is the current real-time available capacity of the transformer, which serves as a hard constraint for all capacity expansion operations. The calculation principle involves dividing the total available capacity by the power required for a single charging station to operate.

[0146] However, the calculation does not simply use the maximum rated power of the charging pile (such as 7kW or 60kW) as the denominator. A more scientific and safer "equivalent rated power" is used for calculation. This equivalent value is equal to the nameplate rated power of a single charging pile multiplied by a "derating factor" less than 1. The derating factor takes into account several practical factors: First, it considers the reality that not all charging piles will start and operate at absolute full power simultaneously, resulting in a peak-shifting effect; second, it reserves a capacity buffer to cope with natural load fluctuations and calculation errors; in addition, the factor may be fine-tuned based on the current comprehensive evaluation score of the power grid, and if the power grid condition is not optimal, the factor value will be more conservative.

[0147] Therefore, the formula for calculating the maximum expandable access quantity N is: N = Floor(Current Real-Time Available Capacity / (Single Pile Rated Power × Derating Factor)). The calculation result is rounded down to ensure that the access quantity is a safe integer value. This value is also compared with an absolute upper limit, and the smaller of the two is taken. This absolute upper limit is determined by the number of physical connection points in the power distribution system (such as remaining vacant piles), thus ensuring the feasibility of capacity expansion operations from both logical and physical dimensions.

[0148] Step S423: Send a charging pile availability notification to the user terminal. The availability notification includes the number of available charging piles and recommended charging time periods.

[0149] After calculating the scalable capacity, availability information is proactively pushed to potential users—an intelligent service strategy that transforms "passive waiting" into "proactive guidance." The generation and delivery of notifications is a personalized and information-rich process. A structured message is pushed to users' smartphone apps, mini-programs, or other terminal devices through the backend interface of the operations management platform.

[0150] This availability notification is carefully designed to include several key pieces of information to help users make a decision: First, it clearly states the number of available charging stations and their specific locations; second, it provides a strongly recommended charging time, typically based on a forecast of available capacity curves, indicating the window with the most abundant and longest-lasting off-peak capacity, such as "We recommend charging between 11:00 PM tonight and 6:00 AM tomorrow"; in addition, the notification may include economic incentives, such as special discounted electricity prices available for charging during the current off-peak period, thereby further attracting users through price leverage.

[0151] The push strategy also has a certain degree of intelligence. It can selectively push notifications to the most likely user groups based on the user's historical charging habits, the vehicle's current battery level (if vehicle data is shared), or their set preferences (such as "finding low-cost charging"), thereby improving the conversion rate and capacity utilization of notifications.

[0152] Step S424: Receive the charging access request submitted by the user terminal, allocate a charging pile according to the charging access request, and send the allocation result to the corresponding user terminal.

[0153] After sending a notification to the user, the process begins receiving and processing requests. Once the user sees the availability notification through the terminal application, they can submit a charging access request with a single click if they agree to the recommended terms. This request is typically a data packet containing information such as the user ID, desired start time for charging, desired charging amount, or target SOC.

[0154] Upon receiving a large number of concurrent user requests, the system processes the requests and allocates resources according to a set of fair and transparent rules. A common strategy is the "first-come, first-served" principle, which processes requests in the order of their precise timestamps arriving at the server and allocates charging station usage rights to the first N requests (N equals the maximum scalable access number).

[0155] Once a charging station is successfully allocated, an allocation result containing information such as the specific charging station number, authorization code, and reservation time slot is generated and immediately pushed to the corresponding user's terminal via the server. Simultaneously, this allocation information is also synchronized to the charging station controller, marking its status as "reserved" to prevent duplicate allocation. For users who fail to secure capacity, a friendly notification message is sent, informing them of the reservation failure and offering the option to join the waiting queue. Through this series of operations, a complete closed loop is established from capacity awareness, calculation, notification to final resource allocation, achieving efficient, fair, and automated allocation of charging resources during off-peak hours.

[0156] Accordingly, please refer to Figure 2 A second aspect of this invention provides a dynamic load balancing device for an intelligent charging pile management system, which performs dynamic load balancing processing on intelligent charging piles based on the aforementioned dynamic load balancing method for the intelligent charging pile management system, including:

[0157] Data acquisition module 1 is used to acquire the operating parameters of the target transformer in the current control cycle of the intelligent charging pile management system, so as to obtain the real-time total load of the target transformer in the current control cycle.

[0158] Data calculation module 2 is used to dynamically calculate the real-time available capacity of the transformer based on the rated power and real-time total load of the target transformer area.

[0159] Prediction calculation module 3 is used to calculate the predicted available capacity of the transformer in the next control cycle based on historical power data and external environmental variables through a load prediction model.

[0160] The equipment balancing module 4 is used to dynamically adjust the operating power of several connected charging piles based on the real-time available capacity and the predicted available capacity, and to expand the number of connected charging piles during off-peak periods.

[0161] Accordingly, a third aspect of the present invention provides an electronic device, including: at least one processor; and a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform the above-described dynamic load balancing method for a smart charging pile management system.

[0162] Accordingly, a fourth aspect of the present invention provides a computer-readable storage medium having computer instructions stored thereon, which, when executed by a processor, implement the above-described dynamic load balancing method for the intelligent charging pile management system.

[0163] The embodiments of the present invention aim to protect a dynamic load balancing method and device for an intelligent charging pile management system, which has the following effects:

[0164] 1. By introducing a forward-looking capacity calculation and dynamic scheduling mechanism based on a multivariate prediction model, the utilization efficiency of power grid resources has been significantly improved. It not only calculates the current available capacity based on real-time transformer load, but more importantly, it integrates historical power data, environmental variables, and user reservation information to accurately predict the load demand and predicted available capacity for future periods through a load prediction model. This enables the system to proactively identify off-peak periods and intelligently expand the number of accessible charging piles while ensuring continuous safety. This solves the problem of resource idleness caused by the lack of prediction capabilities in traditional systems and maximizes the potential of existing power grid infrastructure.

[0165] 2. By monitoring multi-dimensional operating indicators such as grid frequency deviation, voltage imbalance, harmonic distortion rate, and load fluctuation in real time, the system dynamically calculates and adaptively adjusts the safety margin, replacing the traditional conservative strategy of fixed margin. Combined with pulse width modulation technology, the system performs orderly and precise power adjustment on the connected charging piles, forming a closed-loop control system that can sensitively respond to the real-time health status of the grid. This fundamentally enhances the robustness of the system, effectively prevents overload risks, and reduces unnecessary capacity waste.

[0166] 3. By integrating three-phase load balancing optimization functions, the overall stability of the system and the lifespan of equipment are improved; the current and voltage parameters of the three-phase output terminals of the transformer in the distribution area are monitored in real time. Once the imbalance exceeds the threshold, the optimization program is automatically started. The optimal power distribution scheme of each phase is solved by establishing a linear programming model, and precise power redistribution operation is performed to ensure that the three-phase load tends to be balanced. This effectively avoids safety hazards such as low efficiency and equipment overheating caused by single-phase overload, and ensures the stable and efficient operation of the power system.

[0167] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0168] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0169] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0170] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0171] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A dynamic load balancing method for an intelligent charging pile management system, characterized in that, Includes the following steps: Obtain the operating parameters of the target transformer in the current control cycle from the intelligent charging pile management system, so as to obtain the real-time total load of the target transformer in the current control cycle. Based on the rated power of the target transformer area and the real-time total load, the real-time available capacity of the transformer is dynamically calculated; Based on historical power data and external environmental variables, the predicted available capacity of the transformer in the next control cycle is calculated using a load forecasting model. Based on the real-time available capacity and the predicted available capacity, the operating power of several connected charging piles is dynamically adjusted, and the number of available charging piles is expanded during off-peak periods.

2. The dynamic load balancing method for the intelligent charging pile management system according to claim 1, characterized in that, The calculation of the predicted available capacity of the transformer within a future set time period based on historical power data and external environmental variables using a load forecasting model includes: Obtain a multi-dimensional load prediction feature set, which includes: historical load sequence data, external variable sequence data, and scheduled load sequence data; By using the attention mechanism integrated in the load prediction model, the weights of each sequence data in the multi-dimensional load prediction feature set are dynamically calculated and weighted and fused to obtain the future load sequence of the total load prediction value of the target transformer area in the next control cycle. Based on the rated power of the transformer and the predicted load values ​​in the future load sequence, the predicted available capacity corresponding to the next control cycle is calculated.

3. The dynamic load balancing method for the intelligent charging pile management system according to claim 2, characterized in that, The process involves dynamically calculating and weighting the data sequences in the multi-dimensional load prediction feature set using the attention mechanism integrated into the load prediction model, thereby obtaining the future load sequence of the total load prediction value of the target transformer area in the next control cycle. This includes: The historical load sequence data, external variable sequence data, and scheduled load sequence data in the multi-dimensional load prediction feature set are normalized respectively to obtain standardized multi-source time series data. The standardized multi-source time series data is input into the attention calculation layer of the load prediction model. The attention weight distribution of each sequence data at different time steps is obtained by calculating the similarity between the query vector and the key vector. The query vector is generated from the hidden state of the current time step, and the key vector is generated from the time step features of each sequence data. Based on the attention weight distribution, the value vector is weighted and fused to generate a comprehensive feature representation. The value vector is generated from the feature representations of each sequence data. The comprehensive feature representation is input into the time-series prediction layer of the load prediction model for processing to obtain the future load sequence of the total load prediction value of the target transformer area in the next control cycle.

4. The dynamic load balancing method for the intelligent charging pile management system according to claim 2, characterized in that, The historical load sequence data includes: the total active power value sequence, the total reactive power value sequence, and the three-phase current imbalance sequence of the target transformer area within the historical time window; The external variable sequence data includes: the predicted environmental temperature sequence for the next control cycle, the predicted solar radiation intensity sequence, and the date type identifier sequence, wherein the date type identifier is used to distinguish between weekdays, weekends, and statutory holidays; The reserved load sequence data includes: a sequence of expected charging power values ​​for confirmed charging reservations, a sequence of reservation start times, and a sequence of estimated charging durations.

5. The dynamic load balancing method for the intelligent charging pile management system according to claim 1, characterized in that, Before dynamically calculating the real-time available capacity of the transformer based on the rated power of the target transformer area and the real-time total load, the method further includes: Obtain a multi-dimensional set of operational evaluation parameters for dynamically assessing grid stability and load volatility; Based on the multi-dimensional operation evaluation parameter set, the dynamic safety power margin adjustment value of the transformer is calculated; The real-time available capacity is calculated based on the rated power and the real-time total load, combined with the dynamic safety power margin.

6. The dynamic load balancing method for the intelligent charging pile management system according to claim 5, characterized in that, The multi-dimensional operation evaluation parameter set includes: grid frequency deviation, three-phase voltage imbalance, total harmonic distortion, load standard deviation, and maximum mutation rate; The calculation of the dynamic safety power margin adjustment value of the transformer based on the multi-dimensional operation evaluation parameter set includes: The power grid frequency deviation, three-phase voltage imbalance, total harmonic distortion, load standard deviation, and maximum mutation rate are normalized to obtain the corresponding normalized values. The normalized value is multiplied by the corresponding preset weight coefficient and then summed by weight to obtain a comprehensive evaluation score that represents the overall operating status of the power grid. The comprehensive evaluation score is linearly mapped to a preset percentage range, and the mapping result is used as the dynamic safety power margin adjustment value. Wherein, the lower limit of the percentage range is a first preset percentage value of the rated power of the transformer, the upper limit is a second preset percentage value of the rated power of the transformer, and the comprehensive evaluation score is positively correlated with the dynamic safety power margin adjustment value.

7. The dynamic load balancing method for an intelligent charging pile management system according to any one of claims 1-6, characterized in that, Expanding the number of accessible charging stations during off-peak hours includes: When the predicted available capacity is continuously greater than the real-time available capacity and exceeds a preset threshold, it is determined that the load is in a low period. Calculate the maximum number of charging piles that can be expanded to be connected based on the ratio of the current real-time available capacity to the rated power of a single charging pile. Send a charging pile availability notification to the user terminal. The availability notification includes the number of available charging piles and the recommended charging time period. The system receives a charging access request submitted by the user terminal, allocates the charging pile according to the charging access request, and sends the allocation result to the corresponding user terminal.

8. The dynamic load balancing method for an intelligent charging pile management system according to any one of claims 1-6, characterized in that, The dynamic adjustment of the operating power of several connected charging piles includes: The total power adjustment is determined based on the difference between the real-time available capacity and the predicted available capacity. Based on the preset charging pile priority rules, the power adjustment sequence and adjustment amount for each charging pile are determined; A control command corresponding to the adjustment amount is generated using pulse width modulation technology; The control commands are transmitted to the target charging pile via high-speed power line carrier.

9. A dynamic load balancing device for an intelligent charging pile management system, characterized in that, The dynamic load balancing method for the intelligent charging pile management system as described in any one of claims 1-8 performs dynamic load balancing processing on the intelligent charging pile, including: The data acquisition module is used to acquire the operating parameters of the target transformer in the current control cycle of the intelligent charging pile management system, so as to obtain the real-time total load of the target transformer in the current control cycle. The data calculation module is used to dynamically calculate the real-time available capacity of the transformer based on the rated power of the transformer in the target area and the real-time total load. The prediction calculation module is used to calculate the predicted available capacity of the transformer in the next control cycle based on historical power data and external environmental variables through a load prediction model. The device balancing module is used to dynamically adjust the operating power of several connected charging piles based on the real-time available capacity and the predicted available capacity, and to expand the number of connected charging piles during off-peak periods.

10. An electronic device, characterized in that, include: At least one processor; The system includes a memory connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, cause the at least one processor to perform the dynamic load balancing method for the intelligent charging pile management system as described in any one of claims 1-8.

Images